A curated directory of documented incidents where AI-generated and vibe-coded software failed in production. Every entry cites its authoritative source.
Last updated: March 2026
34
Incidents
6.3M+
Records affected
35+
CVEs tracked
69
Vulns found
34 incidents
6-hour outage wipes 99% of U.S. order volume
Incorrect delivery times appear in shopping carts
Claude Code runs terraform destroy, nukes 2.5 years of production data
Compromised PyPI package deploys credential harvester on 95M monthly downloads
Zero-click hack hijacks BBC reporter's laptop in live demo
MoltbookCRITICALFeb 2026Misconfigured database exposes 1.5M auth tokens and 35K emails
CVE-2026-0755: critical command injection, CVSS 9.8
AI agent posts incorrect security guidance, grants unauthorized access for 2 hours
2,000+ vulnerabilities and 400+ exposed secrets across vibe-coded apps
CVE-2026-31992: allowlist bypass via env -S, CVSS 9.9
Agent destroys entire project by looping move command to non-existent directory
AI agent deletes and recreates environment, causes 13-hour outage
IDEsaster: 30+ flaws and 24 CVEs across every major AI IDE
AI agent deletes 200+ emails from Director of AI Safety
IDE forks recommend malicious extensions from unclaimed namespaces
94+ unpatched Chromium vulnerabilities affect 1.8M developers
AI-generated malware exploits Next.js for pre-auth RCE, compromises 91 hosts
CVE-2025-53773: wormable RCE via prompt injection, CVSS 7.8
DNS data exfiltration leaks API keys and secrets from developer machines
126 malicious npm packages exploit AI hallucinated package names
Replit AI agent violates code freeze, wipes entire production database
72,000 images and 1.1M private messages exposed via open Firebase bucket
Malicious code injected into official VS Code extension release
MCPoison: MCP trust bypass allows persistent malicious command execution
CVE-2025-48757: missing Row Level Security exposes 170+ apps
Rules File Backdoor: hidden Unicode instructions weaponize AI assistants
Startup shuts down after AI puts all security logic on the client side
Broken access controls expose every app on the platform
CVE-2025-55284: DNS exfiltration of secrets via prompt injection
CVE-2025-59944: case-sensitivity bypass leads to RCE, CVSS 8.0
CVE-2025-54135: remote code execution with no user interaction
AI-hallucinated package name gets 30,000+ real downloads
CISA-flagged: AI preview workflows expose developer local files to network
Claude scaffolds multiplayer game with instant RCE via insecure pickle serialization
These failures share a common root cause: code was shipped by people who did not understand it. AI generated something that looked correct, passed a cursory check, and went to production. The result was exposed databases, lost orders, and vulnerabilities that required zero user interaction to exploit.
The pattern is accelerating. CVE entries attributed to AI-generated code jumped from 6 in January 2026 to 35+ in March. A Tenzai study found 69 vulnerabilities across 15 apps built by 5 major AI coding tools. Every single app lacked CSRF protection. Every tool introduced SSRF vulnerabilities.
The antidote is the same as it has always been: understand your code. Data structures, algorithms, system design, and the ability to reason about what software is actually doing. AI is a powerful tool when wielded by someone who understands the output. Without that understanding, it is a liability.

Why Vibe Coding Won't Replace Developers
1.7x more bugs, 2.74x more vulnerabilities, 19% slower. The full data-driven analysis with 14 sources.
Read the article ↗